a quick you're probably not going to be taking pictures of the slides either because they're not going to be quite relevant to some of the stuff that you're doing but hopefully we'll get there and we'll we'll work out why i've been invited anyway or you can maybe maybe that can

The Challenge of Being There for People

be what you can tell me at the end so hands up for anyone who's ever felt like they needed to be more places than one that they thought they could be more helpful in more places than one yeah i get that and also uh if anyone's ever felt like it would be helpful for them to be able to be cloned you know if i could be if there was just more of me see i like you again now yeah Yeah, if we could do that, that'd be something that'd be really helpful.

Now, I wonder if any of you have been in a situation where you have been trying to help somebody but didn't know what to say or you've maybe had somebody trying to help you and you could tell that they didn't quite know what to say and you're finding that difficult.

Now, I'm not going to ask you to all get vulnerable with that one too much, but I know from my side of it as a mental health professional, I can find that really uncomfortable. I can find that plays on my insecurities, it can play on me feeling not good enough, It can play on the fact that if I'm receiving that support, then I'm being misunderstood. I'm not quite being misunderstood in that moment. We'll come back to that.

A Personal Journey Toward AI for Good

So I'm Andy Barlow. As you've seen, you've seen everything that you're not supposed to see on there, basically, aren't you? This is a behind the scenes. So I'm going to try and share this journey with you about how we're trying to implement AI. AI for good.

I'm not a tech developer. I'm not even doing a great job at being able to put together some slides But we're going to get there

So, the principle behind it I have struggled with addiction for many years I managed to lose pretty much everything Not have access to it from my son House, home, business, finances, all of those things Which, I know what you're thinking That sounds like a really good Tinder profile No commitments, that's what comes up there No commitments

Now, I've been in recovery for the last eight years, so now I live a clean life by choice. What I learned along that journey has been that I felt misunderstood because people didn't understand what was going on for me, nor did I, to be honest with you.

I met a lot of people along the journey of recovery as well and doing what I do where they felt that when they were being supported, the people that were supporting them didn't quite have the right answers. they didn't have the right information they didn't they weren't prepared majority of people that are supporting somebody with a mental health problem or have got that themselves are not registered nurses they're not professionals they don't have these skills they've possibly gone to google to try and get the answers and that's not that advised or they've gone to chat to try and get the answers and again they don't get the right guidance and these people People are looking for the ways to be able to help.

Learning What Meaningful Support Requires

So what we've been trying to do, 34 years is how long our organisation's been running. I'm second generation. My mum set it up back in 1992. I've been involved since then.

So I've been learning how to speak to people, not how to do slides. I've been learning what the interaction is, looking for pattern recognition, getting a lot of people that come and ask us and say, you know, my friend or my family, they're struggling, I don't know how to help them, what can I do?

So, throughout that time, I've been able to learn that there is a pattern, there are ways in which we can support people better, there are ways in which certain techniques that work, but they are unique. We're from Barnsley, which is just a beautiful place just down the M1, if anybody's wanting to visit, and what we've learnt is that we do things differently in our organisation, you know, we do things quite differently and we get amazing outcomes.

and I've been trying to work out a way in which I can... I've just realised I've put a six -click fantasy. That sounds wrong, that, doesn't it? That sounds very wrong.

From Experience to an Accessible Tool

But the fantasy within that has been the 20 years that I've been a registered nurse, I've been looking at how can I get this message out to more people? How can I help more people to be able to support somebody better?

How can I, if I turn it into a website, then how would that help if you're just about to go in somewhere and they're going to, what, look on a website on their phone? it's just impractical and then i was trying to look at what about if we use youtube and

then people could get the training off of youtube but then it'd be well they're struggling with this particular thing so what is it it's uh series 7 minute 32 it's just it's discombobulated it's difficult it's challenging it wouldn't none of those things would work so what i found with

getting introduced to AI is that there is a way to get to these things much quicker and part of that journey was the fantasy that with six clicks I'll be able to just upload a load of IP I'll be able to give it these prompts in what these previous times I was trying to take notes my

face were looking weird I was trying to take note of right will this help me will I be able to do this just simplify the process of being able to get the information that we've got into somebody's hand to make this application that they will then go oh i can better support somebody better i can

Supporting People the System Has Let Down

understand myself better bless you yeah we we provide complex care and support for individuals whom have lost all their opportunities bridges have been burnt uh they are they struggle maybe on a day -to -day basis of whether they want to be here or not so you can imagine some of the

things that that involves people who have got an inability to regulate their emotions and they'll go from 0 to 10, and that will be screaming and shouting, and lots of places will just say, I'm sorry, we're not going to support you anymore.

So they've lost placement after placement after placement after placement because they've been misunderstood, because they haven't had an opportunity.

The system will say no. If you look at the paperwork, the paperwork would say no, don't give them a chance. It doesn't make sense, unless you do it differently, unless you do it differently.

Why Generic AI Is Not Enough

If you put some of these guys that we support, If you were to use just a general LLM, because the system, the LLM is going to be set up with looking at patterns, it will see that majority of the information it's got is that majority of people should respond in this particular way. So that doesn't allow for the unique ways of approach and to do things. So what we're hoping to do is to be able to help give judgement and give direction in the hands of the people who are actually about to go in and support somebody.

So we're hoping to create something that would help them to be able to go to the next visit, they've heard that this person is going to be difficult to work with, they can click onto their application and they can actually get some guidance and some direction which is fed into by us not just by the LLM it's our bespoke approach all of our IP which is a which

Building a Bespoke AI Approach

has been a big journey in itself so to start off with I've created my own cloud to try and just be recording what I'm doing data protection is a big part of that so I can't put specifics into it which is the really valuable part which is the what we're coming on to one of the biggest challenges

that we've had so I've been able to kind of keep using it keep using it about ideas principles philosophies writing what my approach would be with this type of presentation these type of characteristics so that Claude itself can start to speak more like me not Barnsley but but more like me and not the cookie cutter which is what I was finding because he was constantly trying to move me towards mainstream responses the AI was constantly trying to move me to towards well Well, this is what we've got most research on, but that doesn't mean it works.

A lot of that research will say that being hospitalised, being heavily medicated and losing independence, losing autonomy.

Data Protection and Sensitive Information

So we recognised quite quickly that the sensitive records were going to be a big challenge for the growth and the development and the learning because we cannot put, and still at this stage on the journey, we cannot put all of our client data and stick it into the model.

There are many hoops in which we're having to jump through to be able to get to a point where once we finish this prototype, we will be able to start to use real -life data so that on a case -by -case, we'll be able to grow and develop because that's the way that this will work.

The biggest challenges that we had along the way was the first three months of NDAs and non -disclosures and non -competes because we'd got Claude discussing with a second party and a third party who were also using Claude. So Claude was arguing with Claude about what's actually going to happen and what we should be doing until we exposed the fact that Claude's arguing with Claude.

And I thought I'd broke Claude for a minute. It stopped and went, oops. I was like, oh, and I've not been a techie. I thought, oh, I've just lost everything.

All this work's gone. It's disappeared. And then it just came up with a rational response as soon as it realised it was talking with itself, which was great.

Use a load of tokens in the meantime, obviously. Very happy to do that.

The Reality of Developing a Prototype

So I thought, once we've got all that stuff done, I'm handing it over to the devs. that's going to be dead easy that was the fantasy then we have the prototype we have the sorry we

have the first sessions the development sessions which for anybody that's looking at doing this you know the guys are brilliant they came through to Barnsley we spent the day looking at how we would want it to operate what kind of different techniques we wanted to use what the IP is and

after that session after the second session I thought again you know the fantasy of this thing is going to be that's it now I've done I can just carry on with my work and and they're going to to come back to me with this working thing working prototype that's not what happened i had to load

open source lots of practical and physical ip training sessions create simulated versions of people we've supported to try and get it to come out with the various different responses that

we're needing for it to do and with that what we started to realize is that so much of this is unique and in my head so if there's any operators that are looking at trying to replace themselves and to be like we started off with you in lots of different places then most of what you're doing is probably not going to be thoroughly processed it's not going to be the procedures it probably won't be the policies if you were to operate following all of your policies and procedures on every day you probably wouldn't get a great deal done you know it's learning the the kind of using your intuition to navigate through those different policies and procedures and

And on a day -to -day basis, for us, we're working frontline. I still work frontline in what we do now. And that's where the real learning has been able to come from.

That's been the bit that's hardest and takes more out of me in this journey because the reality of it is it requires me to spend a lot of time, a lot of time.

Testing Decisions Where Lives Are at Stake

I got sent these utterances and there was 150 utterances, which was a kind of an explanation of a scenario that we've put through from the simulated sessions that we've done.

and that would then lead into what would this person do? Who should it be escalated to? And there's various different routes. At what point can we, how can we manage the risk?

Because we're dealing with people, you know, no offence to anyone that's kind of doing finance or tech, but there's lives at the end of this. And we wanted to make sure these people that we support

have already had a really rough deal. They're already struggling and suffering. The last thing we want to do is to start giving them some tech, which they're not going to want to adopt to start off with.

We're not going to want to give the team and the people out in the community tech that's going to end up then giving really shoddy answers or cookie -cutter answers again. We want these to be specific and unique to the individual because the AI that we're creating is going to understand the individual by understanding and learning about their history, by getting all the details.

AI That Guides, Learns and Upskills

The notes that go into this system that we're creating is that every time there's been a visit, before the visit, what will happen is the AI will be able to give guidance of how to handle that visit, give containment to the person that's going into that room to help them to kind of prepare both mentally and also for the technical side of it. During the visit, they'll be able to feed more information into the system, so it will give more guidance in real life whilst the visit's actually happening.

So, my view is that the box ticking of if we don't use this with real life data, then both of the end users, both the client and the person trying to support, are going to find themselves struggling. The quality is going to be lost if we're trying to make this just a tick boxing exercise. The staff and clients are shaping the outcome.

So, we've created governance boards and we've created a professionals board, a clinical board. and client and staff boards which help to answer every single question that we're building into this because the more information from the people in which we're actually going to be supporting and using then the better the quality of the output of what we're providing the governance professionals and the expert boys so the final point on there is that we're hoping to create something that's

going to be able to upskill the staff what we've found so far from technology is it can often and de -skill people sat nav is the great example i used to know all the streets in barnsley or i at least knew one of the main streets in a street off of it now i don't know any of these streets because i get in the car and i tell it where i'm going so it's de -skilled me completely you know going up and down the country i've been de -skilled a lot of technology that i use has de -skilled me

the intention of this is that this is going to be ethical ai that is going to upskill and develop the staff members because it's going to ask them questions it won't just tell them what to do because that's too much direction and that isn't going to be authentic i'll ask them questions about what's the training been doing what things have you done recently it'll guide the staff to

the training it will after the visit it'll give them guidance to the training again to say you asked quite a few questions in these particular areas this will be the thing that would be helpful for you to learn a little bit more about going back to the specifics because it will be the the actual training the minutes of training that are relevant to that particular person

to help and because that's happening through the visit and at the end of the visit we're getting in really valuable data, live data, which is going to be operational in the next person coming to visit that particular person.

From Fantasy to Ongoing Development

So the fantasy was that the dev team had been involved and I'd have little to do and within four weeks we'd have something to work with.

You know, the reality of it, it's required me to put a lot more time and effort into it.

It's allowed me to then, you know, I'm not a tech person, but now somehow I've ended ended up with a GitHub and I've ended up with repositories. I've only known of suppositories and now I know about repositories.

And I'm using cloud code and starting to develop things because I can test it quickly and relatively cheaply as well.

But what I'm very realistic about is these guys are doing an amazing job at building a solid foundation, something that will be able to be multi -tenant. So this can go out to 10, 1 ,000, a million people people potentially, however many, no limit.

What I'm creating, what I'm playing with is something that's kind of in Tornado Alley. It's got no foundation. It can just get blown over and rebuilt and blown over and rebuilt.

But it allows me to keep playing with it and growing and developing, trying to understand a little bit more.

I might even be able to put slides together by the end of the year.

Guardrails, Input Quality and Real-World Practice

The quality of the output, as we've already discussed, is going to be down to the quality of the input, which is the real people putting in the data, the staff members, the clients, not being sat somewhere in an office suddenly writing all these ideas and what might be a good idea. It's what's working in real life.

A Long-Term Vision for Better Care

So return to being out of depth like we started at the beginning. I have had to ask for help and still have to ask for help a lot to be able to turn this into something that's actually going to make a difference

because there are so many carers out there who come into care they are beautiful souls but they are also individuals who've got their own struggles and you bring can't you say you don't bring your home to work but sadly the reality is that that can happen whereas if there's more guidance and we can we can give the support to people in the moment when they need it

not trying to phone the office so i'm going into this visit uh can anybody speak to me with you know there's hundreds of visits happening everybody can't get spoken to at the same time I would love to be able to guide every individual at every visit this might be the way where we can do something like that you know the if we can grow the staff judgment in an area that's already very fragmented and stretched that would be a beautiful thing and that's also for the general public as a

whole um the multi -layer website uh scenario that I put in there is that what we realize is this is very complex to create and to build and obviously these guys I'm not doing it obviously but what

What we're recognising is the process that we're going through is potentially going to be a programme and a process that could help other industries where they have that unique and bespoke element within it that they're needing to do. They're needing something that's more than just being able to produce what would be a simple application.

A Reusable Model for Other Industries

So final note is that thankfully AI is going to be taking our jobs because I don't have the time to do my job anymore because I'm too busy working for AI and getting all the information. So I'll be grateful for that.

The Journey Continues

And the final thing is that I've realised that the more I learn, I thought the finish line was much closer. I thought that it was going to be in four weeks, we'd have a prototype and all the rest of it. The reality is that the finish line was always much further away. It's always much further away. There are so many more steps to take.

And the key for me on this journey is not to give up because I don't like tech. I'm not a tech person, but I can see the immense value of it. And it would be naive to think that this is not the way that we're going to be going. The future of any industry will have the implementation of AI, which is what Nick's here to talk about more.

So cheers, guys.